Issue 76

M. A. Pascal, Fracture and Structural Integrity, 76 (2026) 49-66; DOI: 10.3221/IGF-ESIS.76.04

The FNN-based exponential model predicted accurately with high fidelity to non-linear degradation or empirical corrosion based observations, providing a more realistic risk-informed approach to predicting structural longevity than purely linear extrapolations that do not take into account the varying corrosion rates across components. The model accuracy is evident based on the very small absolute error (typically <0.1 mm) and relative errors (< 1%), even with the highest relative error for Nozzle A1 of 1.25%, and all the errors for F Head 2 were very low, below 0.12%. These differences indicate that the models performed consistently regardless of section size and thickness, making it reasonable to have confidence in the predictive value of the exponential FNN representation for reliably predicting the long-term performance. The linear model, although very simple, has the advantage of low complexity and presents an assumption of constant degradation over time, typically underpredicting or showing losses in future thicknesses (i.e., Nozzle N1, Shell 2), thus bringing into question the risks surrounding reliability-centered maintenance, especially at the minimum thresholds, as it did not include the uncertainty of those predictions. Finally, the exponential FNN showed a statistically significant improvement over the linear baseline, conforming more closely to realistic corrosion degradation patterns.

Figure 7: Thickness predictions for pressure vessel sections.

The anticipated minimum wall thickness trends across various sections of the pressure vessel, as shown in Fig. 7, show the historical data from 2002 to 2008 and the projected data to 2040. The exponential feedforward neural network (FNN) model provides a good fit to the measured data in a non-linear decaying pattern, which is consistent with the measured data and reflects the moderated corrosion rate with the later data range. The linear model implies a linear degradation rate and

58